A single CD4 count at 48 weeks can save lives in HIV patients just as effectively as monitoring every 12 weeks, according to a major analysis of the DART trial. This matters because clinical trials routinely suffer from missing data, flawed statistical methods, and hidden biases that can make their conclusions unreliable. The researchers have developed a new framework for handling missing data in clinical trials, already adopted by a Drug Information Association working group and used in a pharmaceutical company’s regulatory submission. They also tackled the problem of non-proportional hazards—where a treatment’s effect changes over time—which standard survival analyses often mishandle. If these methods become standard practice, regulators and clinicians will have more trustworthy evidence to guide treatment decisions. The work on causal models also allows researchers to extract useful information from trials about second-line treatments and patient management decisions that are normally left to clinician discretion, turning routine trial data into actionable insights. The impact is methodological: cleaner, more honest statistics that prevent flawed studies from influencing medical practice.
View original technical description
We cover issues such as survival analysis methods that allow for non-proportional hazards in both trials and IPD meta-analyses, modelling of prognostic and predictive factors, and analysis of longitudinal and clustered data. On the application of causal models, CTU trials are a rich resource for evaluating aspects of patient management other than the randomised comparison, such as the impact of second-line or concomitant treatments. Although often left to clinician discretion, the trial will usually collect information on the basis for these decisions. Exploiting such data in a major causal analysis of the DART trial in HIV, we showed that 24-weekly and 12 weekly CD4 monitoring give similar results, and that a single CD4 count at 48 weeks leads to better survival than no CD4 monitoring. Finally, on missing data, even modest amounts of missing data can lead to bias and make study conclusions unreliable and/or imprecise. Some methods to deal with it can lead to further bias or imprecision, yet prevail in many disease areas, and are recommended by some regulators. In collaboration with the London School of Hygiene and Tropical Medicine (LSHTM), we have proposed a new framework for the analysis of clinical trials with missing data, which has been adopted by a Drug Information Association working group and for a pharmaceutical company regulatory submission.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know